Predictive Modeling of Escherichia coli Growth: The Role of Key Cellular Features

This study investigates the influence of four key features on the added length of Escherichia coli cells using a fully connected neural network (FCNN), based on data collected from ۱,۲۲۰ samples. The data comprises observations of individual cell and ۱۰-minute sliding window averages from simulated data. Results show that removing the feature fluorescence intensity (YFP) led to the highest increase in Loss (۰.۷۱۱) and root mean square error (RMSE) (۰.۶۹۲). Removing cycle duration (Tcyc) also significantly reduced model accuracy, increasing Loss (۰.۲۸۱) and RMSE (۰.۵۰۲). In contrast, eliminating size at birth (Lb) and growth rate (Mu) had less impact. These findings highlight the importance of effective feature selection in predicting cell growth. This study applies a fully connected neural network (FCNN) to analyze how specific cellular features contribute to E. coli growth predictions. Unlike conventional approaches that rely on broad metabolic profiling, this work emphasizes the predictive power of five key cellular features: growth rate (Mu), added length (Dl), size at birth (Lb), cycle duration (Tcyc), and fluorescence intensity (YFP). The dataset consists of ۱,۲۲۰ samples, integrating both experimental observations and simulated data, with ۱۰-minute window averages used to smooth growth fluctuations. We identify fluorescence intensity (YFP) and cycle duration (Tcyc) as the most influential predictors by systematically removing individual features and assessing their impact on model performance. This insight enhances the interpretability of deep learning models in microbial research and has practical implications for bioprocess time fluorescence monitoring could serve as a reliable proxy for predicting E. coli growth in industrial bioreactors. As machine learning continues to shape modern bioprocessing, integrating feature-driven predictive models could improve strain engineering, enable adaptive process control, and enhance overall yield optimization. This study contributes to this growing field by demonstrating how feature selection and deep learning integration can

کلیدواژه ها:

Escherichia coli ، Cell Growth Prediction ، Metabolic Regulation ، Microbial Cell Dynamics ، Deep Learning in Biology